Running switchover may leave some slaves only [Running] and they cannot continue replication. The effect is somewhat random and probably depends on timing between queries during switchover and replication lag. Seems to be fixed by swapping the order of "FLUSH TABLES;" and "FLUSH LOGS;" during master demotion. Another way to fix this would be to add some slack time and more checks and MASTER_GTID_WAIT-calls to make sure the cluster is in sync before "STOP SLAVE".